There’s a scene that repeats in our meetings: someone from leadership, smart and genuinely interested, tells us “we have to do something with AI”. When we ask which problem they want to solve, the honest answer is usually: I don’t know yet, but we can’t fall behind.
That sentence doesn’t describe a strategy. It describes fear. And fear, fed by hype, is making investment decisions in a lot of companies right now.
Where the confusion comes from
It isn’t the decision-maker’s fault. The environment is genuinely confusing:
- Everything is called AI. Vendors of every kind slapped “AI” on whatever they already sold. From autocomplete to classic automation flows, everything gets presented as artificial intelligence today.
- Demos are not products. A spectacular demo on prepared data is easy. A system that works with YOUR data, your exceptions and your real users is a completely different thing.
- The discourse comes in extremes. Either AI will do everything, or it’s all smoke. Neither extreme helps you decide what to do on Tuesday at your company.
The wrong ideas that cost the most
These are the misconceptions we most often see turn into badly invested money:
“AI will replace the team.” What we see working is something else: AI takes repetitive work off people’s plates so they perform better. Whoever buys “replacement” usually ends up with an abandoned project and a defensive team.
“We need our own chatbot / our own model.” Almost never the right starting point. It’s the modern version of “we need an app”: the shape before the problem.
“It’s plug and play.” Generic tools are; for that same reason, their benefit is also generic. Real value appears when AI is connected to your data and your processes, and that is implementation work, not a purchase.
“We must do something NOW, whatever it is.” The urgency is real, but the right answer to urgency is starting well, not starting anywhere. A badly chosen pilot burns budget and, worse, burns the internal trust you’ll need for the second attempt.
“We bought the tool, so we’re doing AI.” A license without use cases, connected data or measurement is an expense, not a capability.
How to decide better
The rules we use with our clients, and with ourselves:
- Start from the problem, not the technology. “We lose hours on X” is a good starting point. “We need to use AI” is not.
- Demand to see the system on your data. Not on the demo dataset. Before signing, a test on one of your real cases.
- One workflow first, measured. Pick a process, define what success looks like, measure before and after. Scale only what proved to perform.
- Prefer the boring wins. Document classification, support triage, reconciliations, first-line responses. They return more than the spectacular project, and they fund the next step.
- Put a senior filter on your side. Someone with nothing to sell you who reads the proposals and asks the uncomfortable questions. It’s one of the jobs of a fractional CTO.
The good news
Under the noise, the technology is real and the moment is real: implementing AI well has never been this accessible. That’s exactly why it hurts to watch budgets burn on the wrong version of it.
If you’re evaluating an AI investment and want an honest read before signing, bring the problem. Thirty minutes, no deck. If our answer is “don’t do it”, we’ll tell you that too.
